Deployment options

CKEditor AI backend is available in two deployment modes: Cloud (SaaS) and On-premises. Both options provide the same core AI features – Chat, Quick Actions, Review, and Translate – as well as MCP support. The on-premises version offers additional capabilities such as custom AI models and providers.

The Cloud (SaaS) deployment offers the fastest way to get started with CKEditor AI. The AI service is hosted and managed by CKEditor, so there is no server-side setup required on your end. You only need to provide a valid license key and a token endpoint, and configure the editor-side plugins as described in the integration guide. The token endpoint is a service in your application that issues the tokens the editor sends with each AI request. See Authentication in the Cloud Services documentation.

For more information about the Cloud AI service, refer to the CKEditor AI Cloud Services documentation.

The on-premises deployment allows you to run the CKEditor AI service on your own infrastructure, including private cloud environments. The service is distributed as Docker images compatible with standard container runtimes.

On-premises deployment gives you full control over the AI service, including the ability to use custom AI models and providers, and to define MCP (Model Context Protocol) servers in the service configuration.

For detailed setup instructions, requirements, and configuration, refer to the CKEditor AI On-Premises documentation.

To point the editor to your on-premises AI service, set the config.ai.serviceUrl property to the URL of your on-premises instance. The editor also needs the URL of your token endpoint in config.cloudServices.tokenUrl:

ClassicEditor
	.create( {
		licenseKey: '<YOUR_LICENSE_KEY>',

		cloudServices: {
			tokenUrl: 'https://example.com/cs-token-endpoint'
		},

		ai: {
			serviceUrl: 'https://your-on-prem-host.com/v1',

			// ... Other AI configuration options.
		}

		// ... Other editor configuration.
	} )
	.then( /* ... */ )
	.catch( /* ... */ );

To build the token endpoint for your on-premises service, see Step 5: Create the token endpoint.

The on-premises version supports custom AI model providers: OpenAI, Anthropic, Google, Azure OpenAI, Amazon Bedrock, Google Vertex AI, and any provider compatible with the OpenAI Chat Completions API (e.g. OpenRouter, Together AI, or a self-hosted model). Models configured on the server side will automatically appear in the editor’s model selector.

For configuration details, refer to the on-premises configuration guide.

Both deployments let administrators register MCP servers through the admin REST API. The on-premises deployment can also define MCP servers in the AI service configuration. Learn more in the MCP support guide.